Project Summary/Abstract Approximately 1 million patients per year receive radiation therapy in the United States as part of their cancer care. Rapidly advancing technology has over the past two decades enabled treatments in which irradiated volumes are made to conform ever more tightly to tumors. Increasing conformality has resulted in better cure rates and lower side effects, but requires highly precise patient positioning with little margin for error. Patient positioning is performed by radiation therapy technologists, usually by visually aligning on-board x-ray setup images to the CT scans used to plan patients' treatments. Academic studies, public records, and works of investigative journalism have demonstrated that, despite quantitative positioning processes and extensive and rigorous quality control, human error leads to so-called never events: treatments with serious alignment errors or with the wrong patient's treatment plan. Based on never events reported at UCLA, at least 1,400 such events occur nationally per year. Though rare, radiotherapy never events have potentially devastating consequences, and reducing their occurrence is strongly motivated. Radiotherapy is already subject to intensive quality control procedures, so further reduction is likely best achieved through automation in order to avoid additional burden on an already labor-intensive workflow. This project will develop an automated, on-line never event prevention system (NEPS) that will interlock the radiotherapy machine to prevent treatment if the patient is not correctly aligned or if the wrong patient plan is loaded, reducing never events by an order of magnitude and directly addressing the AHRQ priority of improving patient safety. Additionally, this project will retrospectively measure the never event rate at UCLA and Veteran's Health Administration (VHA) radiotherapy clinics, testing the hypothesis that radiotherapy never events are significantly under-reported. In Aim 1, a planar x-ray-based never event detection algorithm will be developed, expanding on an existing volumetric-image based never event detection algorithm. It will be shown that the planar x-ray algorithm has a sensitivity of at least 90% and a specificity of 99%. In Aim 2, the automated never event detection algorithms will be retrospectively deployed to measure the actual never event rate, which is hypothesized to be significantly greater than the reported never event rate, by analyzing over 250,000 setup images from existing UCLA and VHA clinical image databases. In Aim 3, the NEPS will be developed and deployed in a UCLA radiotherapy suite and assessed in terms of number of never events detected, false positive rate, costs of system implementation including personnel costs to address false positives, and potential unintended consequences. This Aim will provide important information regarding feasibility for broader dissemination and implementation. If successful, this project will ultimately lead to increased patient safety by significantly reducing radiation therapy errors with little or no increase in treatment cost or complexity.